The role of the complexity of the seismic source in the generation of near-field high-frequency ground motion is explored. To this end, large (20 km with a resolution of 100 m) and complex seismic fractures are generated with a stochastic growth model depending on two parameters. Synthetic accelerograms are then generated in a high-frequency approximation in which only abrupt changes in the speed of the rupture front are considered. Slip is uniform except for a ramp at the rupture front, and the Green function for a homogeneous, infinite medium is considered.
The approximate records are first compared with exact ones and shown to be in excellent agreement for source-observer distances larger than one or two wave-lengths. Examples for different geometries are given that exhibit the dependence of parameters such as peak acceleration, strong-motion duration, and polarization on the observer's azimuth for a given event. Average records are computed for 1000 events on a 100 by 100 grid; average accelerations do not show a strong azimuth dependence, but peak accelerations vary up to a factor of 5 due to distance and directivity effects. Fourier amplitude spectra for acceleration are computed at a distance from the source half the size of the rupture zone fully taking into account its finiteness. Spectra rise linearly with frequency for a step-like rupture and, for a rupture front lasting about 10% of the total rupture time, they first rise linearly and, at frequencies higher than the risetime, flatten out in agreement with the partial stress drop model of Brune (1970) and consistent with an ω−2 fall-off for displacement.